Podcast Radar Cn

Fetch a candidate set first:

python3 {baseDir}/scripts/fetch_xyz_rank.py --list hot-episodes --limit 20

Filter by genre, freshness, or query when the user already has a direction:

python3 {baseDir}/scripts/fetch_xyz_rank.py \
  --list new-episodes \
  --limit 12 \
  --genre 社会与文化 \
  --freshness-days-max 30

If and only if you truly need extra context for a small set of candidates, enrich a few Xiaoyuzhou URLs:

python3 {baseDir}/scripts/enrich_xiaoyuzhou.py \
  --episode-url https://www.xiaoyuzhoufm.com/episode/69bf524c2d318777c9169361

Workflow

1. Classify the request as one of:
   • listener discovery
   • creator benchmarking
   • curation/distribution research
2. Interpret the user's wording before choosing the list:
   • if the user says 热门播客, 最近播客, 值得听的播客, or similar casual Chinese phrasing, default to episode-level results
   • if the user explicitly asks for 播客频道, 播客栏目, 节目主页, show-level, 频道级, or wants a creator benchmark list, use podcast-level results
3. Pick the most relevant ranking list.
   • default listener-discovery wording usually maps to hot-episodes or new-episodes
   • explicit channel / show / benchmark wording usually maps to hot-podcasts or new-podcasts
4. Fetch enough candidates to support filtering; do not stop at the first 5 unless the user asked for 5.
5. Use ranking fields and title signals first.
6. Only if the answer would otherwise be weak, enrich a small set of Xiaoyuzhou pages.
7. Return a task-shaped result, not raw JSON.

Query Interpretation Rule

In Chinese product usage, users often say 播客 when they really mean 最近值得点开的内容.

Default behavior:

• 热门播客
• 最近热门播客
• 最近值得听的播客
• 推荐几个热门播客

Treat these as requests for episode-level recommendations unless the wording clearly asks for show/channel-level objects.

Switch to podcast-level results only when the user explicitly asks for things like:

• 播客频道
• 播客栏目
• 播客节目主页
• 频道级榜单
• 栏目级榜单
• show-level benchmark or channel list

When in doubt:

• listener-oriented wording -> episode-first
• benchmark / channel / host / show-portfolio wording -> podcast-first

Title-First Rule

Before reaching for Xiaoyuzhou pages, inspect:

• title
• podcastName / name
• primaryGenreName
• rank
• playCount / avgPlayCount
• commentCount / avgCommentCount
• freshness fields

The fetch script also extracts title signals such as:

• episode markers like S8E9, EP03, Vol132
• guest hints like A×B, 对话某某
• format hints like 对谈, 访谈, 复盘, 盘点
• topic keywords inferred from the title

If those signals are enough, do not enrich.

Read references/title-signals.md when you need examples or interpretation guidance.

Xiaoyuzhou Enrichment Rule

Xiaoyuzhou enrichment is allowed only for narrow, necessary follow-up work.

Hard rules:

• never bulk-enrich by default
• never enrich just to complete every field
• never enrich more than 20 URLs in one run
• if a request implies short-window bulk access, refuse enrichment and stay at ranking level

Use enrichment when:

• a short list needs better recommendation reasons
• you need the real Xiaoyuzhou pid from an episode page
• you need a podcast brief or episode description for a handful of finalists

The enrichment script enforces the cap for you.

Output Modes

Use one of these result shapes:

• Listener Discovery: what episode is worth hearing now, and why
• Creator Benchmarking: which shows are worth studying or comparing against
• Curation and Distribution: which shows or episodes are worth packaging, recommending, or developing into downstream ideas

Read references/output-modes.md when you need concrete formatting guidance.

Data Caveats

• 中文播客榜 is weekly, not minute-by-minute real time
• the data is not full-platform coverage
• primaryGenreName is often missing on new lists
• openRate / avgOpenRate can exceed 1; treat these as internal ranking signals, not literal rates

Read references/api.md when you need field notes or endpoint details.

Scripts

• scripts/fetch_xyz_rank.py Purpose: fetch and normalize ranking data, with basic filters and title-signal extraction
• scripts/enrich_xiaoyuzhou.py Purpose: enrich a small set of Xiaoyuzhou pages with brief/description/shownotes context, under a strict cap

What To Avoid

• dumping raw fields without synthesis
• pretending the ranking is a real-time universal truth
• using Xiaoyuzhou enrichment for large-page traversal
• over-explaining openRate as if it were a clean probability